A hybrid method combining Markov prediction and fuzzy classification for driving condition recognition

Driving condition adaptive control is an effective vehicle fuel-saving technique, and the key challenge lies in improving the recognition accuracy of current driving condition. The state-of-the-art approach is based on recognizing historical driving data with a fixed length sliding window to detect current driving condition. However, few research has been conducted to directly recognize the occurring micro-trip (a speed time series between two starts). That is because at the beginning stage of an occurring micro-trip, its known speed time series is too short to be correctly recognized. In this paper, we addressed this issue by proposing a hybrid method for the occurring micro-trip recognition, and two efforts are made to improve recognition accuracy. First, a hybrid recognition procedure is proposed, which combines the Markov chain prediction model and the fuzzy classification model. Second, a statistic approach is proposed to estimate the best time to switch between above-mentioned two models to achieve higher accuracy in detecting current driving condition. Our evaluation results on real-world driving data show that our proposed solution has better accuracy than the state-of-the-art approach.